Stuff The Internet Says On Scalability For November 15th, 2013

Hey, it's HighScalability time:

Test your sense of scale. Is this image of something microscopic or macroscopic? Find out.

  • Quotable Quotes:
    • fidotron: It feels like we've gone in one big circle, where first we move the DB on to a separate machine for performance, yet now more computation will go back to being done nearer the data (like Hadoop) and we'll try to pretend it's all just one giant computer again.
    • @pbailis: Building systems from the ground up with distribution, scale, and availability in mind is much easier than retrofitting single-node systems.
    • @merv: #awsreinvent Jassy: Netflix has 10,000s of EC2 instances. They are the final deployment scenario: All In. And others are coming.
    • Edward Capriolo: YARN... Either it is really complicated or I have brain damage
    • @djspiewak: Eventually, Node.js will reinvent the “IO promise” and realize that flattening your callback effects is actually quite nice.
    • @jimblomo: A Note on Distributed Computing advised against treating remote calls as local functions, I advise against treating the cloud as datacenter
  • Instagram: A Brief, Rapid History of Scaling Instagram (with a tiny team). Have a bigger team. Do simple first. Reduce moving parts. Test and monitor everything. Use Chef earlier. Match the team with the tools. Simple isn't always hacky. Rocketship scaling has been somewhat democratized. 
  • Most of your hard drives will live longer than a mouse (4 years) says Backblaze, the mad scientists of mass storage.
  • Good overview of Day 1 of the Amazon's reinvent conference. And some tweets. New High I/O SSD instance types. More stuff is happening between regions. An RDSized PostgreSQL is now available. As PostgreSQL has some cool geosearching ability this is a nice option for many.  Desktop as a Service. AWS pricing flywheel explains much: behind the numerous price reductions that has AWS has seen. And the wheel starts with external infrastructure innovation that drives to lower infrastructure costs, that allow AWS to reduce prices, that makes AWS attractive to more customers, which then use AWS more, that then needs to buy more infrastructure, enabling economies of scale and the whole process repeats.
  • Mark Earls on Herding Behaviour: The underlying structure of most social networks that we’re interested in is not hub-and-spoke as the influential hypothesis suggests but much looser; and most social networks in the modern world are fluid, rather than fixed. similarly, most social influence is mutual, rather than one way. It’d be nice (and certainly much neater) if things were structured the way the Influentials gang describe but most evidence from broader studies suggests otherwise.
  • A fun article on curiosities related to A Dozen Questions about the Powers of Two. I've never had many questions about the powers of two, but if I had them, I'd hope these would be them.
  • Yes, it's true, the stuff inside your stuff is also complex enough to require complex software to run because not everything can be done in hardware. The second operating system hiding in every mobile phone. Engage recursive turtle analogies. 
  • Watch out. If you are offering an infrastructure type of service Amazon is going to eat you. Eventually. Amazon WorkSpaces. Amazon Kinesis.
  • is running on the Play Framework, Akka and Scala.
  • The story of how Facebook handles system configuration for over 17K systems with 4 people. They chose Chef. 
  • Why the future is NOT RESTful: After watching Matt DeBergalis's meteor talk, it confirmed what I have been thinking about REST for a long time. REST is not fit for the next generation of smart client applications because it has not been designed for smart clients.
  • Are I/O latencies killing your performance?: almost 42% of all the servers in the survey had average tempdb data file write latency of more than 20ms, and just over 12% of all servers had average tempdb data file write latency of more than half a second per write. What can you do: move to SSD; Investigate the I/O subsystem: Pathing/network issues, Incorrect SAN settings, Multiple users of the portion of the I/O subsystem, Only having a single tempdb data file; reduce the usage of tempdb: Spill warnings in query plans, Incorrect, excessive usage of temp tables, Index rebuilds, Using one of the flavors of snapshot isolation and allowing long-running queries.
  • Remove your CDN as a single point of failure. Complex, but interesting. Load balancing multiple CDNs or how jsDelivr works. Each file originates in our Github repository. Pull requests for new files are automatically uploaded via SFTP to my origin. jsDelivr uses Cedexis to load balance. Register all providers or platforms and enter the correct CNAME for each. jsDelivr always selects the faster possible provider for each individual user.
  • Etsy has a great filmography of talks online.
  • A bigger picture is emerging. Mesosphere: Rather than provision and manage virtual machines, Mesosphere's data center operating system makes executing and managing applications in a data center just like launching applications on a smartphone by running them on Apache Mesos.
  • Scuttlebutt - peer-to-peer replicatable data structure, intended to be subclassed into a variety of data-models: CRDT, replicatable array, collaborative text-editing, news feed, single value replication, memory capped with expiring keys. This is real science: Efficient Reconciliation and Flow Control for Anti-Entropy Protocols